压电陶瓷驱动器的滑模神经网络控制

由于压电陶瓷驱动器的迟滞非线性严重影响其定位精度,本文提出了一种滑模神经网络控制方法来改善它的性能.用径向基函数神经网络的输出作滑模控制的等价控制量,由迟滞补偿器估计控射器参数误差、外部扰动和近似计算所造成的不确定量对神经网络的输出控制量进行补偿,从而使驱动器系统状态保持在滑模平面上.基于Lyapunov稳定性理论推导了控制器和补偿器的自适应调节律,分析了控制系统的收敛性和稳定性,以可变幅值的低频三角波为参考位移量对控制系统进行了实验测试与分析,结果表明,只采用神经网络控制时的平均定位误差为0.43 μm,最大误差为0.77 μm,而采用滑模控制方法对神经网络控制量进行补偿后,平均定位误差减小为0.27 μm,最大误差减小为0.49μm,定位精度有了显著的提高.

Neural network control for piezo-actuator using sliding-mode technique

As the positioning precision of piezo-actuators is always severely deteriorated by hysteresis nonlinear effect,this paper proposes a neural network control scheme with a hysteresis compensator based on sliding-mode technique to improve the performance of the piezo-actuators.A Radial Basic Function Neural Network(RBFNN) was developed as a equivalent control value in the sliding-mode control and the hysteresis compensator was used to estimate the lumped uncertainty caused by the varying parameters in the RBFNN,external disturbance and the approximate algorithm to compensate the output signal of the RBFNN.For the above steps,the dynamics of actuator was guaranteed on the sliding surface.The adaptive tuning laws of the network and the compensator were derived on the basis of Lyapunov stability theory,and the convergence and stability of the control system were proved theoretically.A low frequency triangle reference displacement with a variable amplitude was used to detect and analyze the effect of the proposed control method.Experimental results show that the mean and maximal positioning errors by the tradition neural network are 0.43 μm and 0.77 μm respectively,but these errors can be reduced to 0.27 μm and 0.49 μm under the sliding model controller.Finally,the positioning precision is approved evidently.